I think I pretty much did it very manually.. not physically coloring in
those pixels, but - pretty much, created some of the structuring elements
which I think that would work better for this particular image.
First I thresholded the image so that its pretty decent and not too thin.
I didn't want to smooth it out using some of the morphing functions because
it seems like there is not enought pixels for this image to do so.
(and the result looked a little bit funny as well - maybe I didn't find the
right ones). And, if I thicken the image, the holes in 'e' or 'a' gets
closed. Basically, I crop a few samples from the image, cleaned it a
little bit, and bwmorph it to get the thined version, then erode the
image with the structural element. From that, I will know the point
at which the image match, thus proceed to color it.
For some of the characters, I tried the suggestion from the class to help
me eliminate some of the unwanted matches. That is, to remove out parts that
the erosion will get picked up (such as i within an l).
Assumption and Weaknesses:
There is many assumptions that I made for this program. For one thing,
I am given the orientation of my image, so that I didn't have to worry
about rotating my structuring element, or in the case of the rotated
image, the maximum amount of degrees to check. I guess, the way that
I made the program work was pretty much a taylored solution. Thus,
if the program is to work for other similar images, it might not
work as well. Things like a given clean image was assumed - therefore
there was no noise removal.
Overall, I think my program is really weak. In that, I pretty much
hand picked the structual element that matches exactly to this
particular image (therefore the particular threshold). But, to an extent,
I think that for image of constant font, and scale, and orientation,
perhaps this might work well.
Possible Future Work:
can't think well right now - but - definitly should figure out a way
to make it a little more automatic, and less adjustment on the programmer's
part.. I think, I probably have forgotten something important.
Original thresholded image
All using the same threshold value: index > 170 &Resulted image
A Closer look at the rotation...
Source Code:
vowel_vert.m: This is the function that generates the color maped image for the vertical images.
vowel_rotate.m: This is the function that generates the color maped image for the rotated image.
read_bw.m: A function that reads the image and thresholded it according to user preference.
coloring.m: A function that colors the image depending on the point, and what to color
